Benchmarking machine learning models for natural disaster prediction with synthetic IoT data

Abstract

Natural disasters pose severe threats to human life and infrastructure, demanding robust early warning systems (EWS) supported by machine learning (ML) and internet of things (IoT)-based sensing. This study benchmarks ML models for predicting floods and earthquakes using synthetic IoT sensor data. A dataset comprising nine environmental and seismic parameters was generated and labeled into three classes: no disaster, flood, and earthquake, where the feature preprocessing was applied during model training. Logistic regression (LR), random forest (RF), and extreme gradient boosting (XGBoost) models were trained and evaluated using accuracy, precision, recall, and F1-score. Experimental results on the World-A test set show that ensemble models consistently outperform LR, with XGBoost and RF achieving F1-scores of up to 97%and99%,respectively, compared to79%forLR.Anindependenttestonthe separately generated World-B dataset revealed that ensemble models maintained higher generalization capability with F1-scores of 80% for XGBoost and 78% for RF. In contrast, LR showed substantial degradation with an F1-score of 54%. Stress testing on the World-B dataset under simulated situations, such as sensor failures, noise injection, and extreme weather events, confirms the resilience performance of ensemble models in comparison to LR. These results demonstrate the usefulness of ensemble learning in handling unpredictable IoT data for disaster prediction and highlight their potential integration into intelligent EWS. Future work will focus on expanding the framework to include cross-time prediction, incorporating additional environmental features, and deploying the models in real-time IoT systems for field validation.